3.6 Benchmark Model Results
3.6.4 Generalized Forecast Error Variance Decompositions
Generalized Forecast Error Variance Decompositions (GFEVDs), developed by Koop et al.[1996],Pesaran & Shin[1998] serve as a useful method of finding out the amount of variability in a variable that can be attributed to itself or other model variables. Similar to GIRFs described earlier, the main advantage of this procedure is that there
25Introduced byChudik & Pesaran[2011,2013].
26For instance, the authors do not include any real variables which may be fundamentally important
for flows and ignore possible cointegration relationships. Furthermore, for flows, they employ a measure that captures only a fraction of actual flows.
27The authors, however, present results with 25% - 75% bootstrap bands, which are less informative
is no need to specify a certain ordering for the variables or countries in the model. On the other hand, given the presence of correlations across model residuals, GFEVDs do not necessarily sum up to one.
Following the above mentioned authors,D´ees et al.[2007a] andSmith & Galesi[2011], Figures A.15-A.19 summarize the results from the GFEVDs for equity flows and debt flows. Precisely, Figures A.15-A.16 present the normalized (to sum up to one) contri- butions of di↵erent fundamentals,averaged across countries. Contributions of domestic and DC model variables are presented separately for each variable. Figures A.17-A.18 document the heterogeneity across countries regarding the importance of di↵erent fun- damentals, which is not possible to see from the average contributions diagrams. Specif- ically, each bar that corresponds to a variable and a rank (eg: 1st or 2nd) represents the total number of times the variable is in the given order in the overall ranking of all (domestic or foreign) variables in terms of their normalized contributions across all sample countries. Finally, Figure A.19 presents the normalized percentage contributions of domestic, foreign flows and DC model variables in explaining PCFs.
[Figures A.15-A.19 in here]
Starting with equity flows, average normalized contributions in Figure A.15 depict that the developments in the local stock markets play an important role in driving flows among other domestic fundamentals. Other important domestic variables are credit rat- ings, real e↵ective exchange rate and inflation. Examining Figure A.17, one can see that in only a few countries real equity prices stands as the most important domestic funda- mental. For a considerable numbers of countries credit ratings, real e↵ective exchange rate and inflation ranks as other top ranking variables in terms of their importance. The results presented imply that there is notable degree of heterogeneity across coun- tries about the importance of di↵erent fundamentals for equity flows. One interesting observation is that real GDP seems not to be a major domestic driver of flows contem- poraneously, but it is important in 4Q contributions. Also, reserves to debt appears to be the top domestic driver of flows for 6 countries in 4Q. Regarding the DC model variables, in the order of importance, real equity prices, V IX and real GDP are the most important DC fundamentals for equity flows.
Figure A.16 and A.18 indicate that there are di↵erences in the importance of underlying fundamentals between equity flows and debt flows. Contemporaneously, real equity prices appears to be the least important fundamental for debt flows, whereas real e↵ective exchange rate, inflation, reserves to debt and real GDP appear as important domestic fundamentals. Among DC variables, the relative ranking of real equity prices,V IXand
real GDP in their importance for debt flows is the same as for equity flows. However, the contribution of real equity prices and real GDP seems to be less for debt flows compared to equity flows, whereas the importance ofV IX for debt flows is notably more than it is for equity flows. This finding is in line with the findings in the previous subsection and with the general perception about the portfolio debt flows being more risky and reversible than portfolio equity flows. In 4Q, interest rates and reserves to debt gain further importance as domestic variables. Examining Figure A.18, the heterogeneity across countries about the relative importance of di↵erent fundamentals seems also to be present for debt flows, as it is observed for equity flows. However, the degree of heterogeneity is considerably more for the importance of pull factors compared to that of push factors.
From GIRFs, V IX emerges as an important determinant for portfolio capital flows, which is in line with the results in this subsection. Unlike the findings of Chudik & Fratzscher (2012, p. 45-46), on average V IX seems to contribute more towards the variability of debt flows than to equity flows in terms of its normalized contribution. Also,V IX seems to be more important than any single domestic fundamental for both type of flows and horizons.
In relation to the long-debated issue of the relative importance of pull and push fac- tors, it turns out to be the case that the latter dominates the former as Figure A.19 depicts. On average, DC variables seem to have contributed towards the variability in PCFs by more than the domestic factors for both types of flows and in both quarters. Partial importance of domestic factors, excluding flows’ own innovations from domestic contributions, seems to increase in the longer horizon of 4Q , even though they are still outweighed by the DC factors.
ConcerningEF?andDF?, which directly proxy for possible inter-linkages of flows across countries, results suggest thatEF?andDF?variables contributes to the variability of their domestic counterparts by more than the domestic fundamentals and almost as much as DC variables on average across countries. Besides the average contributions, there are notable di↵erences across the sample countries in the importance of these factors. Furthermore, there seems to an interesting pattern in terms of the relative importance ofEF?andDF? with respect to DC-push factors. Flows to countries that are smaller in terms of GDP seem to depend more on flows to other countries, especially for equity flows. The correlations of economic size withEF?(DF?) contributions in 0 and 4 quarters are respectively -21% (-01%) and -27% (-05%).28 Furthermore, the correlation
between the economic size of the country and the ratio of normalized contributions of
EF?(DF?) to DC-push factors are -41% (-14%) and -49% (-21%) in 0 and 4 quarters respectively. Hence, the findings imply that PCFs to countries that are smaller in economic size are more subject to spatial dependencies and/or contagion.
In a related GVAR application to net foreign asset positions (NFA) of 3 Latin American countries,Boschi[2007] finds that domestic factors play a greater role for NFA than the external factors, which seems not to be the case for PCFs as the evidence here implies. However, the apparent heterogeneity about the relative importance of variables is also present for the relative importance of domestic and foreign factors, even across types of flows. For instance, the country in which the average forecast errors of equity flows can be explained most by the DC factors is South Korea by approximately 27% in 0Q and 20% in 4Q, whereas the country with least equity flows dependence on push factors is Morocco by 3% in 0Q and 5% in 4Q. Similar heterogeneity of pull-vs-push factors across countries also exists for debt flows; with maximum push-factors contribution of 23% in 0Q and 21% in 4Q for South Africa; minimum of 2% in 0Q and 4% in 4Q for Colombia. Compared to the existing literature, there are several points to be highlighted. First, the evidence fromGhosh et al.[2012] suggest that foreign interest rates are important drivers of flows, whereas the results from the GFEVDs suggest that foreign interest rates are not one of the key drivers of flows. This particular finding is in line withForbes & Warnock [2012a], as the authors find that foreign interest rates are not related to capital inflows surges/stops. On the other hand, Forbes & Warnock[2012a] find that global growth is a key factor for surges and stops, which is consistent with the evidence obtained in here. Regarding the domestic fundamentals, Ghosh et al. [2012] suggests the importance of real e↵ective exchange rate and real GDP whereasForbes & Warnock [2012a] depict some evidence for the role of Y. Although their findings are broadly consistent with the results obtained in this chapter, as discussed above, there is notable degree of heterogeneity across countries for the importance of di↵erent fundamentals. Another interesting result is the relative importance of real GDP for debt flows compared to equity flows (especially in 0Q), which similarly appears among the findings ofForbes & Warnock[2012b] as the authors depict that growth shocks are much more important for debt flows related surges than equity flows related.